MLBuilds with your dataHigh complexityLevel 2 Refined bespoke
Bespoke Churn
A machine-learning model that predicts each customer's likelihood of lapsing from their behavioural, purchase and session history.
A forward risk score that flags who is likely to lapse, so you step in before they go.
Who it's for
Businesses with a recognisable lapse pattern that want to intervene before customers go quiet, not after they have gone.
The engagement
What you get
- Churn-propensity ML model and scoring pipeline
- Churn-risk score attribute written to Braze
- Save / win-back journey recipes banded by risk
- Model monitoring and periodic retraining
What Fuse does
- Define the lapse event and assemble the behavioural feature set
- Train and validate the churn model and calibrate the risk bands
- Deploy the scoring pipeline and write the risk score into Braze
- Design save and win-back journey recipes banded by risk
- Monitor performance and retrain on an agreed cadence
What we need from you
Data & access
- Behavioural, purchase and / or session event history
Your responsibilities
- Provide behavioural, purchase and / or session event history
- Agree the definition of a lapsed customer
- Provide Braze workspace access and a warehouse feed for scoring
Outcomes and proof
- A forward risk score in Braze that recency only hints at
- Save budget aimed at genuinely at-risk, valuable customers
- Retention lift proven with a held-out control on the win-back
Assumptions
- Sufficient history exists to learn a lapse pattern
- A repeatable scoring feed can be scheduled
- Retention offers and creative are owned by the client
Out of scope
- Offer-sensitivity targeting, a separate service that refines who to treat
- Creative and incentive funding for the save journeys